Once in a while I’m in the mood to try AI-assisted, not even vibe coding, just as an experiment — checking the boundaries. On some tasks it’s very fast and feels really good, but sometimes it gets stuck badly, missing steps that are obviously needed. It just hammers away in different random places, “yelling” that it gathered crucial information that explains everything, then reports the issue is fixed. You say, “no, it isn’t.” After several iterations of this, you’re losing patience, so you ask the AI to make a small test harness so it can verify the fix itself. Then it gets stuck in an infinite loop of reasoning, and you end up walking it through the fix step by step, finally resolving the bug. And after 4 hours, you’re thinking it probably would’ve taken you 30 minutes to fix it yourself.